AirCaption

AirCaption is an AI transcriber tool. A Mac and Windows desktop app that transcribes audio and video on the local machine with NVIDIA, Cohere or Whisper models, then lets you edit, translate into 80+ languages and export SRT, VTT or burned in captions.

AirCaption is a desktop application for Mac and Windows that turns audio and video files into captions, transcripts and subtitles. The core promise is privacy and speed: speech recognition runs on the local machine using NVIDIA, Cohere or Whisper models, so files are not uploaded to a server before processing. The site states this explicitly, noting the app works entirely offline out of the box and that nothing leaves the computer unless the user chooses it. The workflow is built around four stages. Media is dragged in and a model generates the first pass of captions. An editor then provides hotkeys, find and replace, and a waveform view for fixing wording and timing. Translation into more than 80 languages is available as an optional step, and finished work is saved as SRT or VTT files, a plain text transcript, or video with burned in captions. Existing SRT and VTT files can be imported for editing, and batches of files can be dropped into a queue so whole folders are transcribed without manual attention. The intended audience is broad but practical: video editors searching raw footage and subtitling final cuts, podcasters producing blog transcripts, journalists working through interviews and press briefings, legal professionals transcribing proceedings, researchers analysing recordings, event organisers improving accessibility, course creators and marketers adding captions for engagement and search visibility. Where AirCaption differs from browser based transcription services is the trade off between local processing and cloud convenience. Desktop software avoids upload time, file size ceilings and per minute server costs, but it depends on the hardware and requires an install. Cloud translation is an opt in feature: it sends only the caption text selected, not the media, which gives a middle path for users who want more languages or faster turnaround on translation while keeping the source files local.